77 EAs, 32 months, one pattern.

What was run

77 unique files after removing duplicates, all .ex4, all run individually in MetaTrader 4 Strategy Tester. EURUSD H1, Feb 2024 - Oct 2026, spread 17 points (1.7 pip), measured on the live feed, Open prices only, $10,000 start.

What came out

1. Most of the batch never traded

34 files opened nothing in 32 months. A large share of what is sold as an "EA" is an indicator in an .ex4 container, a copy-trade endpoint, or a panel. It installs, it shows a smiley face, and it does nothing. This is invisible if your only test is "does it load".

2. 20 opened positions and closed none

One opened 87,580 positions. Another opened 2,954 at a fixed 200 lots. Without a close there is no trade record, no win rate, and no P/L to check — the only artefact is an equity curve, which is precisely what gets screenshotted and sold.

3. Every stop-out is a sizing event, not a signal event

The eight files that hit stop-out all started at 0.01 lots and ended holding between 1.40 and 5.77 lots. The entry logic did not necessarily fail. The position size decided the outcome before the strategy got a chance to.

4. The best performers are the most fragile

Of the five highest raw profits, 4 turn negative once the best 5% of trades are deleted. The largest swing is an EA that finishes at +$6,888 and lands at −$1,878 — an $8,766 dependency on a handful of closes.

5. Three files, one idea

ea066 and ea069 produce byte-identical results: same trade count, same P/L, same lot progression, different file name. With the optimised and cent-account variants, that is one strategy shipped as three products. Anyone who bought two of them bought the same thing twice — and after 32 months it is the only thing in this batch that survived the tail cut on a sample big enough to matter.

What this changes about buying EAs

Ask for three numbers before you pay: how many positions it opened, how many it closed, and what the start-to-peak lot range is. A vendor who cannot answer the first two does not have a track record. A vendor who answers the third with a range that ends an order of magnitude above where it started is selling you a margin call with a delay timer.